ARC-AGI-2 leaderboard
- GPT-6 Astra leads
- 95.0%
- models measured
- 72
- lead over #2
- +0.8 pts
- median
- 52.8%
Leader on ARC-AGI-2 · read October 21
On October 21, 2026, 72 of our catalog's models had a ARC-AGI-2 result, and OpenAI GPT-6 Astra led with 95.0%.
The harder second set of ARC grid puzzles. Share of puzzles solved.
Role in the index
We collect this benchmark and show it as supporting evidence. It is not part of the Vibe Coding Index, because it measures something other than coding and the index is about coding. Read the methodology
| # | Model | Result | Cost per task | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 95.0% | Not published | 88.7 | $50.00 | |
| 2 | GPT-6.1 SolOpenAI | 94.2% | Not published | 88.7 | $10.00 | |
| 3 | Claude Opus 5.5Anthropic | 93.3% | Not published | 92.2 | $20.00 | |
| 4 | GPT-5.6 SolOpenAI | 92.5% | $1.44 | 76.7 | $20.00 | |
| 5 | Gemini 4 ArgonGoogle | 91.3% | Not published | 80.5 | $10.00 | |
| 6 | Claude Opus 5Anthropic | 90.4% | Not published | 81.8 | $25.00 | |
| 7 | Claude Fable 5.1Anthropic | 90.0% | Not published | 89.1 | $50.00 | |
| 8 | GPT-6 SolOpenAI | 89.6% | Not published | 82.8 | $10.00 | |
| 9 | Claude Fable 5Anthropic | 89.2% | Not published | 82.1 | $50.00 | |
| 10 | Gemini 3.8 FlashGoogle | 89.2% | Not published | 70.5 | $4.125 | |
| 11 | GPT-5.5OpenAI | 85.0% | $1.87 | 67.8 | $30.00 | |
| 12 | Gemini 3.7 FlashGoogle | 84.6% | Not published | 69.9 | $3.75 | |
| 13 | GPT-5.5 ProOpenAI | 84.6% | $10.51 | Not benchmarked | $180.00 | |
| 14 | GPT-5.6 TerraOpenAI | 83.9% | $1.09 | 69.0 | $12.00 | |
| 15 | GPT-5.4 ProOpenAI | 83.3% | $16.41 | Not benchmarked | $180.00 | |
| 16 | Gemini 3.1 Pro PreviewGoogle | 77.1% | $0.9622 | 35.3 | $12.00 | |
| 17 | Claude Opus 4.7Anthropic | 75.8% | Not published | 51.1 | $25.00 | |
| 18 | GPT-5.4OpenAI | 74.0% | $1.52 | 37.7 | $15.00 | |
| 19 | DeepSeek V4.1 FlashDeepSeek | 72.9% | Not published | 77.2 | $0.525 | |
| 20 | Claude Opus 4.8Anthropic | 72.1% | $2.7425 | 66.0 | $25.00 | |
| 21 | Gemini 3.5 FlashGoogle | 72.1% | $0.8503 | 42.8 | $9.00 | |
| 22 | Claude Opus 4.6Anthropic | 69.2% | Not published | 66.9 | $25.00 | |
| 23 | Grok 4.6xAI | 67.1% | Not published | 76.4 | $6.00 | |
| 24 | GLM 5.3 FlashZ.ai | 65.8% | Not published | 71.0 | $0.50 | |
| 25 | Grok 4.20xAI | 65.1% | $0.92 | Not benchmarked | $2.50 | |
| 26 | DeepSeek V4 Flash 0731DeepSeek | 61.4% | Not published | Not benchmarked | $1.32 | |
| 27 | Grok 4.7xAI | 61.4% | Not published | 75.0 | $6.00 | |
| 28 | Claude Haiku 5.5Anthropic | 61.3% | Not published | 75.9 | $0.50 | |
| 29 | DeepSeek V4 Pro 0813DeepSeek | 61.3% | Not published | 54.6 | $1.98 | |
| 30 | Claude Sonnet 4.6Anthropic | 60.4% | $2.6966 | 57.8 | $15.00 | |
| 31 | Gemini 3.6 FlashGoogle | 60.4% | Not published | 48.3 | $3.75 | |
| 32 | Kimi K3Moonshot AI | 60.4% | $1.5927 | 61.8 | $11.936 | |
| 33 | GPT-5.6 LunaOpenAI | 59.5% | $0.67 | 61.5 | $1.20 | |
| 34 | GPT-6 LunaOpenAI | 59.3% | Not published | 71.4 | $0.50 | |
| 35 | GPT-5.2 ProOpenAI | 54.2% | $15.721 | Not benchmarked | $168.00 | |
| 36 | GPT-5.2OpenAI | 52.9% | $1.8982 | 58.9 | $14.00 | |
| 37 | Grok 4.5xAI | 52.6% | $0.78 | 75.5 | $6.00 | |
| 38 | Qwen3.8 27BQwen | 42.4% | Not published | 73.6 | $1.35 | |
| 39 | T Inkling SmallThinking Machines | 40.1% | Not published | 29.9 | $1.20 | |
| 40 | Claude Opus 4.5Anthropic | 37.6% | $2.40 | 66.8 | $25.00 | |
| 41 | T InklingThinking Machines | 36.5% | $0.6421 | 49.7 | $4.05 | |
| 42 | Gemini 3 Flash PreviewGoogle | 33.6% | Not published | 75.8 | $3.00 | |
| 43 | GLM 5.2Z.ai | 22.8% | $0.2501 | 57.7 | $7.00 | |
| 44 | GPT-5.4 MiniOpenAI | 18.9% | $0.75 | 28.2 | $4.50 | |
| 45 | GPT-5 ProOpenAI | 18.3% | $7.1432 | Not benchmarked | $120.00 | |
| 46 | GPT-5.1OpenAI | 17.6% | $1.1704 | 53.3 | $10.00 | |
| 47 | Claude Sonnet 4.5Anthropic | 13.6% | $0.7589 | 59.1 | $15.00 | |
| 48 | Kimi K2.5Moonshot AI | 11.8% | $0.28 | 58.5 | $2.50 | |
| 49 | Gemini 3.5 Flash LiteGoogle | 10.3% | Not published | 33.6 | $2.50 | |
| 50 | GPT-5OpenAI | 9.9% | $0.7302 | 72.2 | $10.00 | |
| 51 | o3OpenAI | 6.5% | $0.8339 | 70.7 | $8.00 | |
| 52 | o4 MiniOpenAI | 6.1% | $0.856 | 68.6 | $4.40 | |
| 53 | GPT-5.4 NanoOpenAI | 5.7% | $0.16 | Not benchmarked | $1.25 | |
| 54 | Gemini 2.5 ProGoogle | 4.9% | $0.757 | 37.0 | $10.00 | |
| 55 | MiniMax M2.5MiniMax | 4.9% | $0.17 | 59.4 | $0.95 | |
| 56 | o3 ProOpenAI | 4.9% | $7.5516 | 84.9 | $80.00 | |
| 57 | GLM 5Z.ai | 4.9% | $0.27 | 59.7 | $3.20 | |
| 58 | GPT-5 MiniOpenAI | 4.4% | $0.1977 | 59.8 | $2.00 | |
| 59 | Claude Haiku 4.5Anthropic | 4.0% | $0.3766 | 49.8 | $5.00 | |
| 60 | DeepSeek V3.2DeepSeek | 4.0% | $0.12 | 54.4 | $0.42 | |
| 61 | o3 MiniOpenAI | 3.0% | $0.5472 | 52.1 | $4.40 | |
| 62 | GPT-5 NanoOpenAI | 2.6% | $0.0295 | 34.8 | $0.40 | |
| 63 | Gemini 2.5 FlashGoogle | 2.5% | $0.3191 | 28.7 | $2.50 | |
| 64 | Qwen3 235B A22B Instruct 2507Qwen | 1.3% | $0.0044 | Not benchmarked | $0.55 | |
| 65 | R1 0528DeepSeek | 1.1% | $0.0527 | 71.4 | $2.15 | |
| 66 | GPT-4.1OpenAI | 0.4% | $0.0691 | 46.5 | $8.00 | |
| 67 | Llama 4 MaverickMeta Llama | 0.0% | $0.0121 | 18.1 | $0.6525 | |
| 68 | Llama 4 ScoutMeta Llama | 0.0% | $0.0062 | 9.1 | $0.30 | |
| 69 | GPT-4.1 MiniOpenAI | 0.0% | $0.0139 | 28.5 | $1.60 | |
| 70 | GPT-4.1 NanoOpenAI | 0.0% | $0.0036 | 8.9 | $0.40 | |
| 71 | GPT-4oOpenAI | 0.0% | $0.08 | 38.8 | $10.00 | |
| 72 | GPT-4o-miniOpenAI | 0.0% | $0.01 | Not benchmarked | $0.60 |
Only models in our catalog are listed, and each appears once with its best published run. 197 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out. Cost per task is the spend on one puzzle, in US dollars, as Epoch AI records it beside the result.